DISCUSSION PAPER SERIES

IZA DP No. 14636

Christos A. Makridis

Barry T. Hirsch

The Labor Market Earnings of Veterans: Is Military Experience More or Less Valuable than Civilian Experience?

AUGUST 2021

Any opinions expressed in this paper are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but IZA takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity.

The IZA Institute of Labor Economics is an independent economic research institute that conducts research in labor economics and offers evidence-based policy advice on labor market issues. Supported by the Deutsche Post Foundation, IZA runs the world’s largest network of economists, whose research aims to provide answers to the global labor market challenges of our time. Our key objective is to build bridges between academic research, policymakers and society.

IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.

Schaumburg-Lippe-Straße 5–953113 Bonn, Germany

Phone: +49-228-3894-0Email: publications@iza.org www.iza.org

IZA – Institute of Labor Economics

DISCUSSION PAPER SERIES

IZA DP No. 14636

The Labor Market Earnings of Veterans: Is Military Experience More or Less Valuable than Civilian Experience?

AUGUST 2021

Christos A. MakridisNational Artificial Intelligence Institute at the Department of Veterans Affairs and Arizona State University

Barry T. HirschGeorgia State University and IZA

ABSTRACT

IZA DP No. 14636 AUGUST 2021

The Labor Market Earnings of Veterans: Is Military Experience More or Less Valuable than Civilian Experience?

We assess the labor market experiences of military veterans, focusing on three major

outcomes, among others, controlling for a wide array of demographic characteristics

and industry and occupational fixed effects. First, we find that male and female veterans

receive civilian earnings nearly equivalent to nonveteran men and women. This finding

implies that military experience is valued in the labor market similarly to foregone civilian

experience. Second, veterans are clustered in occupations with somewhat lower than

average employment and real earnings growth, and in metropolitan areas with lower levels

and growth of real GDP per capita. Third, veterans experience lower returns to formal

educational investments (e.g., college) than do nonveterans. Veterans realize earnings

gains from professional licenses, but their returns are lower than for nonveterans. These

gains are concentrated among science, technology, engineering, and math (STEM) jobs,

suggesting that veterans could help meet the growing demand for tech talent and artificial

intelligence skills.

JEL Classification: J3, J4, J44

Keywords: military veterans, earnings levels and dispersion, work

experience, licensing, public sector, occupation growth

Corresponding author:Barry T. HirschDepartment of EconomicsAndrew Young School of Policy StudiesGeorgia State UniversityAtlanta, GA 30302-3992USA

E-mail: bhirsch@gsu.edu

1 I. Introduction

There is a large literature measuring differences in civilian earnings between those who

are military veterans and nonveterans. Many have focused on periods of conflict -- World War

II, the Korean War, and the Vietnam War -- in which a substantial share of those who served did

so because of the draft.1 Several papers have focused on the draft lottery implemented toward the

end of the Vietnam War. Shortly following the draft lottery, however, the U.S. moved to an all-

volunteer military, which remains today.

The two-sided nature of selection into military service affects our measurement and

interpretation of civilian earnings differentials between veterans and nonveterans. In particular,

the current all-volunteer military generates lower dispersion in skills than does a military with a

widespread draft and required military service. In the left-tail of the “quality” distribution, the

military (employer) does not accept recruits with either low ability (test scores), health

limitations, or criminal records. High ability young men and women in the right tail of the

distribution are less likely to enter the military given that military compensation is lower than

they can expect to receive in the civilian labor market.

Given that the military has difficulty attracting enlistees from the far-right tail of the

ability distribution, we expect relatively few veterans to have high levels of lifetime wealth.

Veteran households between ages 55 and 64 have a $71,860 lower median net worth than their

nonveteran counterparts ($160,809 versus $232,669) and that 59% of male veterans between

ages 35 and 44 hold credit card debt, compared with 47.7% among nonveterans (Eggleston and

Holder, 2017). Moreover, whereas the labor force participation rate is 65.5% for nonveterans as

of 2018, it was 49.2% for veterans (U.S. Bureau of Labor Statistics, 2019a).2 1 Angrist (1990) and Angrist, Chen, and Song (2011) provide analysis of veteran/nonveteran earnings differences for those exposed to the draft lottery toward the end of the Vietnam War. Prior analyses on Vietnam-era veterans and nonveterans, as well as for earlier periods, include Villemez and Kasarda (1976), Little and Fredland (1979), De Tray (1982); Berger and Hirsch (1983, 1985), Schwartz (1986), Goldberg and Warner (1987), Mangum and Ball (1989), Angrist and Krueger (1994), and Angrist (1998). Relatively recent analyses include Hirsch and Mehay (2003), Davila and Mora (2012), Card and Cardoso (2012), Faberman and Foster (2013), Routan (2014), and Tan (2020). Some of these analyses can reasonably claim that their earnings gap estimates are causal; most cannot. 2 Labor force participation rates among veterans in 2018 was 49%, as compared to 66% among nonveterans. This difference reflects the older average age of veterans than nonveterans. Participation rates are roughly equivalent for veterans and nonveterans when compared within narrow age groups (BLS, 2019a, Table 2A).

2 The primary contribution of this paper is to provide an updated assessment of the civilian

labor market experiences of veterans and nonveterans over the past two decades, well after the

period of mandatory military service. We document new stylized facts about the increasing

exposure of veterans to polarization in the labor market and somewhat lower returns to education

than seen for nonveterans. While our results are not fully causal, we control for a wide array of

demographic characteristics and industry and occupation fixed effects. We focus on 2005

through 2018, capturing an era with substantial technological change (e.g., Autor and Dorn,

2013; Autor 2019).

The initial focus of the paper is on civilian wage differentials between veterans and

nonveterans, conditioned on education and other correlates of productivity. Using the Current

Population Survey (CPS), we identify many differences between veterans and nonveterans, but

consistently find near-zero estimates of conditional wage differentials, suggesting that time spent

in military service provides an increase in civilian productivity nearly equivalent in value to that

from civilian work experience. When we use data from the American Community Survey (ACS),

however, we find small earnings disadvantages among veterans, although we cannot always

reject the null that their earnings are the same. A positive (negative) veteran-nonveteran wage

differential would suggest that military experience provides human capital and/or other forms of

market productivity that are higher (lower) than are the gains from civilian work experience.

Men’s and women’s veteran status provides a signal to potential employers (Spence 1973). That

said, we find variability with respect to race, ethnicity, gender, and job sector (e.g. public versus

private employment). Taken as a whole, our conclusion is that the value of military experience is

roughly equivalent or slightly below that of civilian experience.3

Our analysis uncovers several patterns not previously examined in the research on

veterans. We find that veterans are concentrated in occupations with lower real hourly wage and

employment growth over the past decade, as well as in metropolitan areas with lower real per 3 Such a conclusion is not unique to recent years. At least one recent paper examines military service during World War I (Tan 2020) and concludes that there was little evidence of a causal relationship between wartime service and subsequent economic outcomes. That said, a recent paper by Gabriel (2020) finds that World War I veterans observed in Census data from 1930, 1940, and 1950 were employed in occupations that provided higher pay and greater upward mobility than did occupations in which nonveterans were employed. Similar evidence is found for World War II, in which there was nearly universal service among men fit for duty (Angrist and Krueger 1994). Men unable to serve in World War II tended to earn less than did their World War II veteran counterparts.

3 capita GDP levels and growth. We show that college-going veterans select a different mix of

majors than do nonveterans. Moreover, veterans with high levels of educational attainment,

either formal (e.g., college) or informal (e.g., certificate/license), exhibit lower returns than their

nonveteran counterparts. That said, the returns to receipt of a certificate/license in a STEM-

related occupation are positive and economically meaningful for veterans. These results offer an

optimistic roadmap for veterans with an interest in the digital economy, especially the emerging

labor market for artificial intelligence related skills. Moreover, since the federal government has

had a tough time retaining and attracting skilled workers (Makridis, 2021), veterans may have a

unique role to play here, particularly with the ease of obtaining additional security clearances.

The structure of the paper is as follows. Section II summarizes our data and sample

construction. Section III investigates earnings differences between veterans and nonveterans

across various partitions. Section IV focuses on gender-specific evidence on veteran-nonveteran

earnings differences. Section V compares results using the American Community Surveys (ACS)

with those obtained using the Current Population Surveys (CPS). Section VI documents new

evidence of polarization among veterans across occupations and metropolitan areas. Section VII

examines veteran-nonveteran differences in the benefits from occupational licensing and

certification. Section VIII examines veterans’ returns to formal schooling, with the focus on

college, college majors, and post-graduate degrees. Section IX explores veterans’ employment

and earnings in the public sector. Section X concludes.

II. Data and Measurement

Our analysis provides descriptive and regression-based evidence on the relative earnings

of veterans and nonveterans. Our primary analysis is based on the monthly Current Population

Survey (CPS) Outgoing Rotation Group (ORG) files for the years 2005-2018. The CPS-ORG

files provide quarter samples that report weekly earnings and hours worked, in addition to the

standard CPS information on demographics, veteran status, household geographic location,

industry, occupation, and other measures. We deflate all nominal variables using the personal

consumption expenditure index normalized to 2012.

Additional analyses include use of recent years of the CPS that provide information on

the presence of worker licensing and certification. We also conduct additional robustness

analyses using the annual American Community Surveys (ACS), which has large sample sizes

4 and includes information on college majors, information not provided in the CPS. Finally, we

also use the Occupational Employment Statistics (OES) from the Bureau of Labor Statistics

(BLS). The OES provides precise measures of both employment and earnings growth at the 6-

digit occupational level, although we aggregate to the 3-digit level.4

Both the CPS and ACS allow us to see whether respondents have served in active duty in

the U.S. armed forces. Veterans are not asked whether they were an enlisted member or an

officer in the armed forces. Officers account for less than one-sixth of those in the military;

hence, our estimates of veteran-nonveteran earnings differentials are heavily weighted toward

enlistees. To be an officer, one must have completed a four-year degree and, upon entry into the

military, commit to a five-year service period. Given these requirements, restricting the sample

to those 26 years old or less can produce a sample entirely of enlisted veterans (we thank a reader

for the suggestion). The downside of doing so is that such a sample is rather small and provides

information only for earners recently exiting the military. Such a group is weighted

disproportionately by veterans currently in school and/or with low earnings unrepresentative of

lifetime earnings. As we subsequently note in the paper, veterans tend to have low earnings

shortly after leaving the military, while catching up over time. We choose to rely on a broad

sample that permits us to identify multiple earnings patterns for veterans and nonveterans.

Our estimates reflect a sample weighted heavily toward veterans who were enlistees, with

a much lower weight on those who were officers. Using administrative data on members in the

Component Reserves, some of whom are veterans and others not, Hirsch and Mehay (2003,

Table 2) provide veteran-nonveteran wage gaps separately for those enlisted and those who were

officers. For those enlisted, the veteran-nonveteran wage gap was effectively zero. For officers,

the veteran-nonveteran wage gap was positive (about 8 percent).

III. Veteran-Nonveteran Earnings Differentials: Heterogeneity with Respect to Race,

Ethnicity, Education, Urban/Rural Residence, and Gender

We begin by estimating earnings regressions for wage and salary workers of the form: 4 With the exception of the OES data on employment and real wages and the Bureau of Economic Analysis (BEA) data on real per capita GDP, all data files were accessed from the Integrated Public Use Microdata Series (IPUMS) at the Minnesota Population Center.

5 𝑤𝑖𝑡 = 𝑉 𝑇𝑖𝑡 + 𝜙 𝑖𝑡 + 𝜉 𝑉 𝑇𝑖𝑡 × 𝑖𝑡 + 𝑋𝑖𝑡 + 𝜖𝑖𝑡 (1)

where 𝑤 denotes the natural log of real hourly wage (deflated to 2012 prices) for individual i in

year t. 𝑉 𝑇 denotes an indicator for whether the individual is a veteran. denotes indicators for

whether individuals are part of a given group (e.g., race, ethnicity, rural area), thus providing a

compact and convenient way to explore heterogeneity in veteran-nonveteran log earnings

differentials with respect to worker demographics and differences in attributes. 𝑋 denotes a

vector of other individual controls. While we also introduce two-digit industry and occupation

fixed effects to account for otherwise unmeasured worker/job skills and working conditions, we

recognize that there are relevant variables not available. Our goal is to document conditional

correlations within comparable jobs. Standard errors are heteroskedasticity-robust.

Our principal results for men are documented in Table 1. As evident in column (1) of the

table, the “raw” (i.e., no controls) log hourly earnings gap of 0.123 indicates a 13.1 percent

earnings advantage for male veterans relative to nonveterans.5 This earnings advantage for

veterans is driven primarily by the older age of veterans than nonveterans. The analysis provides

an assessment of the earnings differentials between today’s male veterans and nonveterans,

controlling first for detailed demographic and educational characteristics, followed by the

introduction of controls for occupation, industry, and time fixed effects.

Although these specifications lack an explicit causal interpretation since we do not

account directly for selection into the military, column 2 of Table 1 indicates that veterans earn

about 1% (0.013) more than their nonveteran counterparts, conditional on demographics

(education, potential experience, race and ethnicity, marital status, and metropolitan size bins).

Once we control for occupation, industry, and time fixed effects (column 3), veteran-nonveteran

wage differentials are effectively zero (0.002). This implies that veterans tend to be employed in

occupations and industries with slightly higher earnings than seen for nonveterans. Although

these industries and occupations tend to have somewhat higher earnings, we subsequently show 5 Throughout the paper we treat the log differential as the percentage differential, albeit one with an intermediate base. The standard conversion from a log differential to an arithmetic percentage is [exp(β)-1]100, where β is the log gap. The 0.123 veteran coefficient implies a 13.1 percent arithmetic differential. Wage gap estimates with controls are far below 0.123; hence there is a minimal difference between the log gaps and the arithmetic percentages.

6 that occupations and metropolitan areas populated by veterans tend to have had slower growth in

hourly earnings and employment.

In columns 4 through 7 of Table 1, we examine heterogeneity in veteran wage gaps with

respect to education levels, race, ethnicity, and urban/rural residence. In column (4), we find that

among those without a B.A./B.S. or above, veterans earn 1.8 percent more than their nonveteran

counterparts. Among those with a college degree, however, male veterans earn 3.7 percent less

than do nonveterans (i.e., .018 – .055). As will be shown subsequently, we find the same pattern

for women veterans. Given the two-sided selection into the military, these results are consistent

with expectations. Veterans are underrepresented in both the far left and right tails of the

earnings distribution. Conditioned on those with college degrees, veterans are expected to rank

below the average for that group. Standard least square results are likely to provide reliable

estimates of veteran wage gaps among those near the middle of the earnings distribution;

estimates may be less reliable in the lower and higher tails of the distribution.

Our results are comparable in several ways from Card and Cardoso (2012) who use

detailed administrative data from Portugal to examine the effects of compulsory military service

on wages. Although they find an overall weak relationship between conscription and wages, they

provide evidence that conscription had a positive effect on wages among less-educated workers.

Consistent with the Portuguese result, we find that less-educated U.S. veterans (e.g., those

without college attainment) receive positive returns to military service.

A complementary explanation for veterans’ lower-than-average returns from a college

degree is that veterans are more likely than nonveterans to have received college degrees from

low-quality for-profit colleges. For-profit colleges have actively recruited veterans; they have

strong incentives to do so. For-profit colleges depend heavily on federally funded student aid.

Under federal law, college revenues from federal monies (mostly student aid) were capped at 90

percent of total revenue. Prior to that federal policy, some of the larger for-profit national chains

such as University of Phoenix and Kaplan University received more than 80 percent of their

revenues from Title IV student aid. The federal Title IV program limits, however, do not include

veteran education benefits toward that 90 percent limit. Hence, for-profit colleges have actively

recruited students with military benefits – primarily veterans, service members, and family

members of those currently in the military (see Deming et al. 2012; Cellini and Goldin 2014).

7 There is rather limited evidence on the returns to education among those attending for-

profit college. That said, Cellini and Turner (2019) used administrative data from the Department

of Education and the Internal Revenue System that enabled them to measure employment and

earnings among those who attended for-profit colleges, as compared to matched students

attending nonprofit institutions. Their findings were clear-cut. They found that certificate-

seeking students at for-profit institutions are less likely to be employed (1.5 percentage points

lower). Among those employed, those from for-profit institutions earned about 11 percent less

than students who had attended public institutions. Outcomes were particularly poor for students

who attended for-profit colleges that were multi-campus chains and those focusing on online

courses. They found similar qualitative results for both men and women and for seven of the ten

most populated fields of study. Cellini and Turner (2019) did not identify students who were

veterans or nonveterans. Given that veterans are disproportionately enrolled at for-profit

colleges, their results provide a likely explanation for our observation of lower returns (wage

gains) to college degrees for male and female veterans.6

Columns (5) and (6) examine veteran earnings differentials with respect to race and

ethnicity. African-American veterans receive roughly 2.5% higher earnings than their nonveteran

counterparts (conditioned on education, occupation, industry, etc.). Although the difference is

modest, it indicates positive selection of black men entering the military, reflected in AFQT

scores, absence of a civilian criminal record, and good health. Moreover, employers are likely to

view veteran status as a positive signal, a la Spence (1973), with respect to expected productivity

and reliability (Spence 1973, Berger and Hirsch 1985). An older literature on African American

veterans found larger veteran-nonveteran wage differentials among black men than what we

observe with more recent data. Although the positive selection story remains relevant, it has

lessened importance over time.7 6 Credentials matter in the larger labor market. Deming et al. (2016) provide results from an experimental study in which fictitious resumes were sent to real vacancy postings. Among their findings were that applicants listing business degrees from for-profit colleges had substantially lower callback than from applicants from nonselective public institutions. An exception was that in health jobs, for-profit credentials coupled with a government occupational license did not have a negative signal. 7 Our analysis focuses exclusively on those employed. There have been considerable reductions in labor force participation and high rates of institutionalized black males (Hirsch and Winters 2014; Bayer and Charles, 2018). If one were to account for African Americans out of the labor force and institutionalized (i.e. zero earners), we would see a larger left tail with respect to skill, thus observing even stronger positive selection by the military.

8 Positive selection also applies to Hispanic veterans. We find a substantial veteran wage

advantage on the order of 10 percent (0.098), four times the size of the veteran-nonveteran wage

differential among black men. The large veteran wage differential for male Hispanics is likely a

combination of both strong skills among Hispanic veterans and relatively weak skills among

many Hispanic nonveterans. The latter group is likely to include those with poor English skills

and/or absence of legal residence (non-citizens must be legal immigrants to enlist in the

military). All Hispanic veterans must be either U.S. citizens or documented; a substantive share

of nonveteran Hispanics are not documented and face substantial employment barriers in the

labor market. Hispanics not documented are less likely to be fluent in English and may lack other

skills valued in U.S. labor markets. In short, veteran status appears to function as a high-valued

credential for Hispanics. Moreover, the U.S. Hispanic population is disproportionately

concentrated in the south and southwest (in particular, Texas and California), as are U.S. military

bases. It is possible that employers in these markets place particularly high value on employees’

military experience, especially so if they conduct government contracting.8

Young men and women growing up in non-urban rather than urban areas are more likely

to enlist in the military. Using CPS data, we can identify households residing in metro areas of

roughly 100,000 and above (roughly 70% of the CPS samples). As is widely recognized,

earnings rise substantially with respect to labor market size, although urban areas have become

much more polarized (Autor, 2019). In column (7) of Table 1, we compare veteran-nonveteran

earnings differentials in non-urban areas to those in metro areas. We had no prior expectation

regarding this relationship. Veteran-nonveteran wage differentials, however, are found to be

about 3 percent (-0.028) lower in non-urban than urban areas. A plausible explanation is that

there are relatively large shares of veterans residing in non-urban labor markets, possibly 8 As seen above, we infer that there is positive selection into the military for both black and Hispanic veterans. In a recent paper from Hamermesh et al. (forthcoming) using time-use data (i.e., the ATUS), the authors measure racial and ethnic differences in “non-work” time during paid time work hours. Both black and Hispanic workers are found to have somewhat higher levels of “non-work” during paid work hours. These differences in “non-work” time on the job explain a relatively small share of the overall racial and Hispanic wage gaps. The authors note, however, that among male veterans, there were no difference in “nonwork” hours between black veterans and white workers, nor any differences in nonwork among Hispanic veterans and white workers in the ATUS samples. In short, the racial and ethnic differences in “nonwork” found by the authors is fully driven by nonveterans. Time-use evidence in this study reinforces our conclusion that there is positive selection into the military among black and Hispanic men.

9 mitigating their value at the margin.9 Alternatively, the veteran signal to employers may be less

important in small markets where personal information on job candidates’ strengths and

weaknesses are more readily available or known.

Our primary results include the entire 2005-2018 period. An obvious question is whether

veteran-nonveteran wage gaps have varied much over time. Given the similarity of our general

results to a much earlier literature, our expectation is that changes over time have been modest.

When we examine veteran wage gaps across three sub-periods, 2005-2009, 2010-2013, and

2014-2018, we find similar veteran wage gaps. For example, for specification (2) for males the

veteran coefficients varied between 0.013 and 0.026. For specifications (3) through (7), veteran

coefficients for each of the three time periods were near zero for all specifications.

Our analysis clearly shows that on average, male veterans receive civilian earnings highly

similar to the earnings of nonveterans. It does not follow, however, that armed service personnel

leaving the military immediately receive civilian earnings equivalent to what they would receive

had they not served in the military and, instead, built up civilian work experience. A RAND

study from Martorell et al. (2013) examined administration data from the Army that followed

veterans after they left the military. As one might expect, they find a substantial dip in earnings

when they first separate from the military, but the veterans on balance did as well or better than

those without military experience. Such a pattern is not unique to a single place or time. Berger

and Hirsch (1983) concluded that Vietnam-era veterans initially suffered earnings penalties after

leaving the military, but subsequently caught up. A more recent paper by Angrist et al. (2011),

using administrative data that followed Vietnam-era veterans, confirms this result. Nor is such a

pattern specific to U.S. veterans. Card and Cardoso (2012) find that veterans in Portugal briefly

face low civilian earnings when they first leave the military, but quickly catch up and/or overtake

the earnings of nonveterans. We are reluctant to conclude that this is a universal result. Recent

evidence from Bingley et al. (2020), who exploit random assignment in the Danish draft lottery,

find a negative effect of peacetime military service on earnings arising from disruptions in

educational investments and resulting careers. 9 The share of veterans in rural areas is higher than in CPS-designated metro areas. Likewise, controlling for standard demographics, we find that male veterans are 6.4 percentage points less likely to live in metro areas with over 1 million people than their nonveteran male counterparts.

10 IV. Do Veteran-Nonveteran Earning Gaps Differ for Women and Men?

Previous analyses of labor market earnings of veterans have focused almost exclusively

on men. Enlistment of women into the military has increased substantially over time, but there

has been limited study of female veterans in the economics literature. A comparison of the

earnings of female veterans relative to nonveterans is of some importance given the substantive

share of women in the military. As of 2019, women made up roughly 20 percent of the Air Force

and Navy, 15 percent of the Army, and 9 percent of the Marine Corps (Dever, 2019). We provide

analysis of female as well as male veterans. We find modest differences between male and

female veteran outcomes in civilian labor markets, but the general patterns seen for men and

women are broadly similar.

Given the increase of women in the military over time, in Table 2 we examine veteran

wage gap estimates for women over the same 2005-2018 period seen for men in Table 1 (the

layout is identical to that seen for men). We do not have strong priors that civilian earnings

among women veterans differ substantially from those of female nonveterans. As seen

previously for men, wage gaps between veterans and nonveterans are rather small, implying that

veterans’ time in the military is valued similarly to the equivalent time spent in the civilian labor

market. Prior evidence in Mehay and Hirsch (1996, Appendix) shows that among early cohorts

of women in the military (those born prior to about 1950), female veterans had civilian earnings

that exceeded those among female nonveterans. Among cohorts born after 1950, however,

female veterans and nonveterans tended to have similar wages.10

Based on those earlier findings, our expectation is that veteran wage gaps for female

veterans are close to zero, just as we have seen for men. Table 2 provides estimates of veteran-

nonveteran log earnings differentials for women, with the structure identical to our previous table

for male veterans. The raw veteran-nonveteran log earnings gap for women (column 1) is 14

percent (0.143), similar to that seen for men (0.123). Focusing on column 2 of Table 2, the

earnings regression with individual controls but not occupation and industry fixed effects, we 10 Mehay and Hirsch (1996, Appendix) provide supplementary analysis of female veterans using the 1989-1993 CPS. For birth cohorts of women between 1955 and 1973, they find wage ratios of female veterans to nonveterans to be very close to 1.0. For much earlier cohorts of women, the civilian earnings of female veterans substantially exceeded the earnings of nonveteran women, but the differences were fully accounted for by standard worker attributes (schooling, age, etc.).

11 find a 4½ percent (0.044) earnings advantage for female veterans. Once one controls for

occupation and industry fixed effects (column 3), the veteran effect is effectively zero. As found

previously for men, college-educated female veterans earn less than do college-educated

nonveterans (specification 4). Hispanic female veterans earn roughly 6 percent (0.063) more than

Hispanic nonveterans, the same qualitative pattern found for men (see the previous discussion),

albeit smaller in magnitude for women. No wage difference (0.001) is found for similar black

female veterans and nonveterans. Female veterans in rural (non-metro) areas earn slightly less (-

0.012) than female nonveterans, the same pattern found previously among men.

V. Veteran-Nonveteran Earnings Gaps from the American Community Surveys (ACS)

We next compare veteran-nonveteran earnings differentials from the CPS with those

from the annual American Community Survey (ACS) over the same years. The ACS has the

advantage of larger sample sizes, providing greater variation and external validity; see, for

example, the U.S. Congressional Budget Office (2017) for its relative advantages. The CPS

measures usual weekly earnings and usual weekly hours worked on the current primary job,

whereas the ACS only provides annual earnings in the prior 12 months, as well as their

occupation and industry. Because the ACS includes weeks worked during the previous year and

hours worked per week, we can control for both as a proxy for the intensive margin.

In the left-side of Table 3 (columns 1-7), we report regression results from the ACS for

male veterans and nonveterans, structured similarly to those seen using the CPS. But there are

differences. The dependent variable using the ACS is the log of annual earnings, in contrast to

the CPS measure of hourly earnings on the primary job at the time of the monthly surveys.

Overall, the ACS results tend to provide similar, but somewhat lower (more negative)

veteran-nonveteran earnings differentials than does the CPS. This is not surprising given

differences in the earnings measures (hourly for the CPS and annual for the ACS). As seen in the

CPS, there is a substantial raw earnings advantage for male veterans of 12 percent (0.123), as

compared to the smaller 7 percent (0.073) annual earnings advantage of male veterans in the

ACS. Both these “raw” veteran-nonveteran earnings measures drop sharply once individual

controls are included (most notably years of potential experience, highly correlated with age),

from 12 percent to 1½ percent with individual controls in the CPS. The comparable ACS

comparison for annual earnings is a decline from the raw 7 percent veteran advantage (column 1)

12 to a minus 2 percent (-0.017) with controls (column 2). We have also found that controlling for

the intensive margin, such as part versus full time work, explains meaningful variation in

earnings between veterans and nonveterans and reduces the differences.

As previously reported using the CPS, we found that the earnings boost from a college

degree among veterans is lower than for nonveterans. The ACS result is qualitatively similar but

shows a larger veteran disadvantage. Both the CPS and ACS show that veteran-nonveteran

earnings differentials are higher for African American than for white workers. Likewise, the

veteran advantage among Hispanic (relative to white) men is substantial, 9 percent higher in the

CPS and 8 percent using the ACS.

The right-hand side of Table 3 (columns 8-14) provides ACS veteran-nonveteran

estimates for women. Focusing on raw veteran-nonveteran gaps based on hourly earnings in the

CPS, we find a 14 percent (0.143) veteran log wage advantage. As expected, using the ACS we

observe a larger 22 percent (0.217) veteran log annual earnings gap. Once we add controls for

demographics (including schooling and potential experience), as well as occupation and industry

fixed effects, we find that female veteran-nonveteran earnings gap using the ACS are effectively

zero. This pattern is identical to what we found previously using the CPS.

A minor issue in our analysis is the presence of imputed earners in both the CPS and

ACS, which causes a slight attenuation of the veteran-nonveteran earnings gap estimates (Hirsch

and Schumacher 2004; Bollinger and Hirsch 2006). Nonrespondents are assigned so-called

allocated (i.e., imputed) earnings from a respondent who has attributes “similar” to the

nonrespondent. In both the CPS and ACS, veteran status is not an attribute included in the

imputation procedure. As a result, among the subset of imputed earner observations (i.e.,

nonrespondents), minimal earnings differences between veteran and nonveterans are reported. In

work not shown, we estimated earnings differentials using respondent-only samples, as well as

full samples including both respondents and nonrespondents. We find differences in the veteran

coefficients between the two samples in the third decimal place, approximately a 0.002

attenuation. In short, veteran-nonveteran wage differences reported in our tables are slightly

smaller (in absolute value) due to inclusion of imputed earners.11 11 The CPS earnings files from IPUMS do not include allocation (imputation) flags for the early years in our sample.

13 VI. Do Veterans Take Jobs with Faster or Slower Growth in Earnings and

Employment?

As seen in our wage analyses, veteran-nonveteran wage gaps are close to zero, and

possibly negative if we defer to results from the larger sample in the ACS. Motivated by the

evidence showing that many veterans do not earn as much as nonveterans in civilian labor

markets, we investigate their employment choices and job matches with respect to occupation

and location. Specifically, we examine the correlation between the shares in veteran status and

the growth rate of occupational employment and real hourly wages between 2002 and 2018, as

well as the correlation with real per capita GDP levels and its growth rate, to understand whether

the patterns of polarization seen in the overall labor market also apply to veterans.12

We conclude that the answer to this question is yes. Veterans are clustered in occupations

with lower employment and real wage growth, as well as in metropolitan areas with lower per

capita GDP and lower per capita GDP growth. Not only is this finding novel, it is also important

for two reasons. First, given the substantial investments in veterans during their service period,

the finding that veterans often enter lower-growth occupations following their service suggests

that military investments are not as effective as they otherwise might be. Given that personnel

payments and benefits account for 39% of the total $686 billion military budget as of 2020, there

are potential areas for efficiency gains.13 Second, if armed service members are facing an

increasingly tough transition into the civilian sector, attracting capable and interested individuals

will become more difficult. Difficulty in attracting individuals into the armed services could have

substantive implications for national security and readiness.

Veterans are employed in occupations that have had lower than average growth in both

employment and earnings between 2004 and 2018. Much of the productivity growth since the

1960s has been concentrated in occupations with high levels of information technology (IT)

intensity (Gallipoli and Makridis 2018). The young men and women who enter the military

disproportionately tend to have aptitude and skills that are broadly average, with relatively few

individuals in the extreme left and right tails of the economy-wide skill distribution. Employment 12 Our results are robust to the base year we use, but we choose to use 2002 as our starting point since it is after the technology boom and bust and before the run-up to the sub-prime mortgage crisis. 13 https://comptroller.defense.gov/Portals/45/Documents/defbudget/fy2020/FY20_Green_Book.pdf

14 and wage growth over time, particularly during the 1990s, has been weakest in the middle of the

occupational skill distribution. More specifically, technological change has decreased demand

for jobs (occupations) that have tasks that are routinizable and programmable. (e.g., Autor 2019,

Autor et al. 2003, Acemoglu & Autor 2012). These labor market changes have

disproportionately harmed veterans. Similar economy-wide changes have occurred in other

developed countries; the average across European Union (EU) countries exhibits occupational

hollowing that is highly similar to that seen in the U.S. (Goos et al., 2009).14

To confirm this pattern, we examine veteran employment shares among occupations and

labor markets. In Panel A of Figure 1, we show the relationship between the share of veterans

employed in 3-digit occupations, averaged over the 2010-2017 ACS, and the annual real hourly

wage growth of these occupations between the periods 2002-03 and 2017-18 from the

Occupational Employment Statistics (OES). Panel B of Figure 1 shows the relationship between

the veteran shares (as above) and the OES annual average employment growth rates in 3-digit

occupations (aggregated from 4-digit data and weighted by employment) over the same period.

The data on employment and wages by occupation is from the Occupational Employment

Statistics (OES) for the years 2002-2018. The simple correlations of veteran shares with respect

to both occupation wage growth and employment growth are -0.12.

In Figure 2, we provide a similar analysis to that of Figure 1, but now we correlate

veteran shares within U.S. labor markets (i.e., metro areas) with respect to metro-level GDP per

capita averages for 2002-2017 (panel A), as well as GDP per capita growth (Panel B).15 The

simple correlations of veteran shares with respect to MSA real GDP per capita is -0.46; the

correlation of veteran shares and MSA real GDP per capita growth is -0.37. These correlations

with respect to metro area GDP levels and growth are substantially stronger than seen previously

with respect to occupation wage and employment growth.

VII. Do Veterans Benefit from Occupational Licensing and Certification? 14 The military has been concerned that many of its soldiers are not well prepared to transition to IT and other rapidly growing civilian occupations. They have supported recent studies guiding veterans toward civilian occupations that utilize and require job skills similar to their military occupations (see Wenger et al., 2017). 15 We would use 2018 data on real per capita GDP to match our real wage and employment data, but it is not available at the time of our release of this paper.

15 There exists a substantial recent literature documenting the increased earnings associated

with occupational licensing (e.g., Gittleman et al., 2018). We explore whether veterans realize

gains in earnings from occupational licensing and certification that are similar to earnings gains

seen in the overall labor market. Beginning in 2015, the CPS added questions asking employed

workers (and those previously employed) if they have an occupational license or certification. In

our January 2015 through June 2019 CPS monthly wage and salary sample, roughly a quarter

(24.4%) of all wage and salary workers have an active professional certification or license. A

subset of licensed workers report whether they have a government-issued certification or license

and whether certification is required for their job. We focus on the broader sample to ensure that

we have enough power for these comparisons, given that veterans are already a small share of

the overall population.

As seen in column 4 of Table 4, we find that the earnings bump associated with licensing

is lower for male veterans than for nonveterans. The average nonveteran male receives weekly

earnings approximately 8-9 percent (0.085) higher than do similar workers absent the license,

conditioned on detailed industries and occupations. The average male veteran, however, receives

a 0.070 licensing bump, 0.015 lower than seen for male nonveterans.16

An identical analysis for women (Table 5) shows that female nonveterans with licenses

receive weekly earnings 8 percent (0.081) higher than do nonveteran women without licenses.

Female veterans with a license, however, receive weekly earnings only 6 percent (0.081-0.022 =

0.59) higher than the earnings of female veterans without a professional license.

In Figure 3, we examine log wage earning returns with respect to licensing for veterans

and nonveterans, estimated separately by broad occupation groups (our sample includes both

women and men). The regressions include the standard set of demographic controls. Overall,

there is minimal evidence for a higher earnings advantage from licensing for veterans as

compared to nonveterans. For most of the occupational categories, there is little difference in

returns to licensing for veterans relative to nonveterans. The one clear exception, however, is the

broad category of math/computer science/engineering and other professional occupations where

we see licensed veteran workers with a 10 percent earnings advantage relative to licensed 16 As seen in column 2 of both Tables 4 and 5 (as compared to columns 3 and 4), the estimated earnings licensing advantage is more than 2 percentage points higher absent controls for detailed industry and occupation.

16 nonveterans. This could capture the fact that the sciences have an overall higher earnings

premium and more efficient means for assessing competency in skills (Kirkeboen et al., 2016).

In summary, the returns to licensing are roughly similar with respect to both gender and

veteran status. That said, we systematically find slightly lower returns to licensing for veterans

versus nonveterans, and lower licensing returns for women as compared to men (see Tables 4

and 5). We previously showed that veterans tend to be employed in occupations that have had

below average growth in both employment and wages. The lower returns to licensing among

veterans than among nonveterans are due in part to the differences in occupational mix, as can be

seen by the decline in the licensing coefficient following inclusion of occupation and industry

fixed effects in specifications (3) and (4) of Tables 4 and 5.

We find these results particularly relevant for the ongoing debate about occupational

licensing reform and inquiry into approaches for better transitioning of veterans into the civilian

sector. For example, our evidence is consistent with the view that occupational licensing aids

veterans to a lesser extent than for nonveterans, perhaps because states have their own licensing

requirements. Veterans and their spouses are likely to relocate across states during and following

active-duty service (Steinhauer, 2020). Moreover, our results offer optimism for veterans who

are interested in STEM jobs, particularly since many are remote in the post-pandemic economy.

The federal government has had a tough time retaining and attracting talent (Makridis, 2021),

especially in STEM jobs and the emerging demand for artificial intelligence. Another reason that

veterans may also be uniquely suited for these jobs is that they will have an easier time obtaining

security clearances having already served in the military, reducing the time it takes to onboard

and work either with federal agencies from the private sector or in them as a civilian.

VIII. Veteran and Nonveteran Returns to College Majors

The occupational mix among college-educated veterans (and nonveterans) is determined

in part by their choice of college majors. Using the 2009-2017 ACS, we compile college majors

for veterans and nonveterans, collapsing what are detailed majors into 9 broad categories, plus a

small “Other” category. A detailed list of majors for each of the broad categories is shown in the

Notes to Figure 4. The compiled results as shown in Figure 4 display the shares of majors for

veterans (shown with red bars) and shares for nonveterans (blue bars). As seen in the figure, the

overall patterns of college majors are roughly similar for veterans and nonveterans, with business

17 majors and social science majors being the two largest categories of study for both groups. Areas

of study in which veterans are more heavily represented are engineering/computer science,

business, the social sciences, and an industrial category (the latter including services in

construction, electrical, mechanical, natural resources services).

To see how labor market returns to college majors differ for veterans and nonveterans,

we examine results from three earnings regressions using the ACS, shown in Table 6. These

regressions provide estimates of earnings by college major for veterans and nonveterans. The

earnings regression shown in column (1) includes no controls, apart from the college majors by

veteran status. Our preferred earnings regression is shown in column (2); it includes a detailed

set of individual controls, but not detailed occupation and year fixed effects.

A third regression, shown in column three, includes the full set of occupation and year

fixed effects. We place less weight on the results from column (3) than on results from column

(2). We downplay the results in column (3) because detailed occupation and industry fixed

effects absorb much of the skill and (to a lesser extent) working conditions that determine

earnings. Inclusion of detailed occupation and industry (i.e., job effects) biases returns to

schooling and other forms of experience (including time in the military). Column 3 in Table 6

shows coefficients that are systematically attenuated, relative to those in column (2).

The nine college major coefficients shown in Table 6 measure the relative returns to each

major among nonveterans, as compared to nonveteran returns in the omitted social science

major. We find nonveteran returns to the education, humanities, industrial/production, and

“other” (non-designated) degrees are lower than nonveteran returns from social science degrees.

Likewise, the returns to nonveterans for communications, engineering/computer science, the

sciences, medicine, and business law majors exceed those of the social sciences.

Our primary focus is on how college returns differ between veterans and nonveterans for

alternative majors. These estimates are shown by the interaction terms of veteran status with

respect to the different college majors. Focusing on column (2), the veteran coefficient is -0.014,

indicating that veterans realize earnings roughly 1.4 percent lower than do nonveterans among

those whose college major was in the social sciences (i.e., the omitted group of college majors).

Most of the coefficients interacting veteran status and a college major/field are negative,

indicating lower returns for veterans than for nonveterans in these fields. Particularly large are

18 the negative coefficients for veteran returns on business/law and education. Working in the

opposite direction, veterans receive higher returns than do nonveterans in four general fields of

study – the sciences, industrial/production, medicine, and engineering/computer science.

IX. Veteran Hiring in the Public Sector

Veterans are disproportionately employed in public sector jobs. This is not surprising. As

discussed previously, veterans are underrepresented in the lower and upper tails of the ability

distribution, while overrepresented toward the middle of the distribution. Veterans’ skills and

experiences often match up well with public sector jobs, be they police, firefighters, paramedics,

nurses, postal workers, or teachers. In addition, there exist veteran preferences in hiring for most

federal jobs (including the U.S. Postal Service). Applicants are rated and ranked by agencies. For

veterans who have an honorable or general discharge, they are typically awarded additional

rating “points” in the ranking of applicants for a given federal position.

Veteran hiring preferences are not restricted to federal jobs. Most state and local

governments also adopt veteran hiring preferences (Lewis et al., 2014). Given that jobs in the

federal government are a small share of all public sector jobs, most public sector workers are

employed by state and local governments rather than the federal government. Public sector

veterans are disproportionately employed in state and local government jobs, particularly so in

male-dominated occupations such as police and firefighters (Lewis et al., 2014).17

Our 2005-2018 CPS sample of wage and salary workers confirms the disproportionate

presence of veterans in public sector jobs. We find 23.4 percent of our male veteran sample

employed in the public sector, as compared to only 12.7 percent among male nonveterans.

Women are disproportionately employed in the public sector. Among our female sample, 31.5

percent are public sector workers, as compared to 19.1 percent among nonveterans.

Given that the public sector is more highly unionized than is the private sector, it follows

that veterans are more likely than are nonveterans to be union members.18 Among all men 17 Based on calculations from the 2018 CPS, 48.8% of public workers are local government workers, 33.6% state workers, 14.6% non-postal federal, and 3.0% postal service. These public workers account for 15.1% of all wage and salary workers. 18Based on the CPS, 33.9 percent of public sector workers were union members in 2018, as compared to 6.4 percent in the private sector (BLS, 2019, Table 3). Roughly half of all union members are public sector workers.

19 (women) in 2018, 11.1 (9.9) percent were union members (BLS, 2019, Table 1). We calculate

that among veterans in 2018, 17.5% of men were union members, while 14.7% of women were

union members.

We showed earlier that veterans are employed in occupations with lower average growth

in employment and earnings. These patterns are reinforced by the concentration of veterans in

the public sector. The share of public sector jobs and wage growth in the public sector have

declined relative to the private sector. Based on CPS data for 2005 and 2018, public sector

employment grew by only 3.6 percent over these years, as compared to 12.8 percent in the

private sector and 11.3 percent for all wage and salary workers. Real hourly wage growth in the

public sector over 2005-2018 period grew by 19.0 percent, as compared to 26.6 percent for the

private sector (Hirsch and Macpherson, 2019, Table 2a).

X. Conclusion

Our analysis provides new evidence on the civilian earnings of military veterans, both men

and women, over the 2005 to 2018 period, based primarily on evidence from the monthly Current

Population Surveys (CPS) and the annual American Community Surveys (ACS). For both men and

women, we find civilian earnings of veterans roughly the same as the earnings of similar

nonveterans. The finding of a near-zero earnings gap has an intuitive interpretation.

While time served in military service decreases time spent in civilian life (e.g., school and

civilian work experience), absence of a systematic earnings gap between veterans and nonveterans

implies that time spent in the military contributes to subsequent civilian earnings roughly the same

as does time spent in the civilian sector. Broadly similar results occur for both male and female

veterans. Selection into the military subsequently leads to a civilian labor force that has a

substantial number of veterans, few of whom have extremely low skills, criminal records, or poor

health due to earlier selection by the military.19 Because earnings are relatively compressed in the

military, high skilled young men and women with high civilian earnings potential are less likely to

enter the military.

Given the military selection process, it is not surprising that we see military veterans

overrepresented in middle-class jobs, while underrepresented in both the lower and higher tails of

the wage distribution. Veterans are less likely to have professional licenses than do nonveterans, 19 The obvious exception are military veterans injured either in combat or in other on-the-job military activities.

20 and their returns to licensing are below those for nonveterans. Veterans are more likely to live

outside of larger metropolitan areas, thus receiving earnings lower than seen in large urban areas.

Because military veterans are highly represented in middle-class jobs, they have been

disproportionately exposed to the hollowing out of many middle-class jobs, due in large part to

technological change.

Veterans have disproportionately entered occupations having lower than average growth

in both employment and earnings. Given these negative forces affecting veterans’ earnings, the

approximate equivalence of veteran and nonveteran earnings could be considered a pleasant

surprise. We cannot predict how future labor market forces will shape the relative earnings of

veterans and nonveterans. It is important, however, that we do not ignore future patterns of

earnings and employment among those who have served in the military.

Although not a major focus of the paper, our results may help guide policy. The

combination of deteriorating outcomes for veterans and low returns to educational investments

point toward the importance of helping service members transition more effectively into the

private sector. To do so, veterans increasingly will need to acquire the skills and/or formal or

informal credentials that enable them to obtain attractive jobs aligned with their skills and

interests. Policy barriers, however, such as state-specific occupational licensing requirements,

can erect barriers that make transitions difficult. Not all veterans are provided with information

about their post-service labor market opportunities in advance, making last-minute efforts too-

little too-late. Applications of artificial intelligence (AI) to individual military experience records

and stated preferences may provide personalized feedback that translates into improved matches

in the labor market. Such endeavors might also provide smoother income streams, as well as

steering veterans toward meaningful work that mitigates the risks of mental health problems

among veterans (Zivin et al., 2011).

References

Acemoglu, Daron and David Autor. 2012. “What Does Human Capital Do? A Review of Goldin and Katz’s The Race between Education and Technology.” Journal of Economic Literature

50(2), 426-63.

Angrist, Joshua D. 1990. “Lifetime Earnings and the Vietnam Era Draft Lottery: Evidence from Social Security Administrative Records.” American Economic Review 80 (3): 313-36.

21 __________. 1998. “Estimating the Labor Market Impact of Voluntary Military Service Using

Social Security Data on Military Applicants.” Econometrica 66(2):249-88.

Angrist, Joshua D., Stacey H. Chen, and Jae Song. 2011. “Long-term Consequences of Vietnam-Era Conscription: New Estimates Using Social Security Data.” American Economic Review:

Papers & Proceedings 101(3): 334-38.

Angrist, Joshua D and Alan B. Krueger. 1994. “Why Do World War II Veterans Earn More Than Nonveterans?” Journal of Labor Economics 12(1):74-97.

Autor, David H. 2019. “Work of the Past, Work of the Future.” American Economic Review:

Papers and Proceedings 109, Richard T. Ely Lecture, 1-32.

Autor, David H. and Dorn, David. 2013. “The Growth of Low Skill Service Jobs and the Polarization of the U.S. Labor Market.” American Economic Review 103(5): 1553-97.

Autor, David H., Levy, Frank, and Murnane, Richard J. 2003. “The Skill Content of Recent Technological Change: An Empirical Exploration.” Quarterly Journal of Economics 118(4): 1279-1333.

Bayer, Patrick and Kerwin K. Charles. 2018. “Divergent Paths: A New Perspective on Earnings Differences Between Black and White Men Since 1940.” Quarterly Journal of Economics 133(3): 1459-1501.

Berger, Mark C. and Barry T. Hirsch. 1983. “The Civilian Earnings Experience of Vietnam- Era Veterans.” Journal of Human Resources 18 (4): 455-79.

Berger, Mark C. and Barry T. Hirsch. 1985. “Veteran Status as a Screening Device During the Vietnam Era,” Social Science Quarterly 66(1): 79-89.

Bingley, Paul, Peter Lundborg, and Stéphanie Vincent Lyk-Jensen. 2020. “The Opportunity Costs of Mandatory Military Service: Evidence from a Draft Lottery.” Journal of Labor

Economics 38(1): 39-66.

Bollinger, Christopher R. and Barry T. Hirsch. 2006. “Match Bias from Earnings Imputation in the Current Population Survey: The Case of Imperfect Matching.” Journal of Labor

Economics 24(3): 483-519.

Card, David and Ana Rute Cardosa. 2012. “Can Compulsory Military Service Raise Civilian Wages? Evidence from the Peacetime Draft in Portugal.” American Economic Journal :

Applied Economics 4(4): 57-93.

Cellini, Stephanie Riegg and Claudia Goldin. 2014. “Does Federal Student Aid Raise Tuition? New Evidence on For-Profit Colleges.” American Economic Journal: Economic Policy 6(4): 174-206.

Cellini, Stephanie Riegg and Nicholas Turner. 2019. “Gainfully Employed? Assessing the Employment and Earnings of For-Profit College Students Using Administrative Data.” Journal of Human Resources 54(2): 342-70.

Davila, Alberto and Marie T. Mora. 2012. “Terrorism and Patriotism: On the Earnings of US Veterans following September 11, 2001.” American Economic Review: Papers &

Proceedings 102(3): 261-66.

Dennis N. De Tray. 1982. “Veteran Status as a Screening Device.” American Economic Review 72(1): 133-42.

Deming, David J., Claudia Goldin, and Lawrence F. Katz. 2012. “The For-Profit Postsecondary School Sector: Nimble Critters or Agile Predators?” Journal of Economic Perspectives 26(1): 139-64.

22 Deming, David J., Noam Yuchtman, Amira Abulafi, Claudia Goldin, and Lawrence F. Katz.

2016. “The Value of Post-secondary Credentials in the Labor Market: An Experimental Study,” American Economic Review 106(3): 778-806.

Dever, Mary. 2019. “With Historic Number of Women in Uniform, the Vet Community Is About to Change,” Military.com, March 11, accessed at https://www.military.com/daily-news/2019/03/11/historic-number-women-uniform-vet-community-about-change.html

Eggleston, Jonathan and Kelly Ann Holder. 2017. “The Wealth of Veterans,” U.S. Census Bureau, Current Population Reports, P70-1, 2017: 70-151, Washington, DC. Accessed at https://www.census.gov/library/publications/2017/demo/p70-151.html

Faberman, R. Jason and Taft Foster. 2013. “Unemployment Among Recent Veterans During the Great Recession,” Economic Perspectives, Federal Reserve Bank of Chicago, 1Q/2013: 1-13.

Gabriel, Paul E. 2020. “An Empirical Examination of the Occupational Attainment of American Veterans of World War I,” Social Science Quarterly 101(3): 1009-17.

Gallipoli, Giovanni and Christos A. Makridis. 2018. “Structural Transformation and the Rise of Information Technology,” Journal of Monetary Economics 97(C): 91-110.

Gittleman, Maury, Mark A. Klee, and Morris M. Kleiner. 2018. “Analyzing the Labor Market Outcomes of Occupational Licensing,” Industrial Relations 57(1): 57-100.

Goldberg, Matthew S. and John Warner. 1987. “Military Experience, Civilian Experience, and the Earnings of Veterans,” Journal of Human Resources 22(1): 61-81.

Goos, Maarten, Alan Manning, and Anna Salomons. 2009. “Job Polarization in Europe,” America Economic Review Papers & Proceedings 99(2): 58-63.

Hamermesh, Daniel S., Katie R. Genadek, and Michael C. Burda. Forthcoming. “Racial/Ethnic Differences in Non-Work at Work, Industrial and Labor Relations Review: 1-21.

Hirsch, Barry T and David A. Macpherson. 2019. Union Membership and Earnings Data Book, Compilations from the Current Population Survey: 2019 Ed., Bloomberg BNA.

Hirsch, Barry T. and Stephen L. Mehay 2003. “Evaluating the Labor Market Performance of Veterans Using a Matched Comparison Group Design,” Journal of Human Resources 38(3): 673-700.

Hirsch, Barry T. and Edward J. Schumacher, “Match Bias in Wage Gap Estimates Due to Earnings Imputation,” Journal of Labor Economics, Vol. 22, No. 3, July 2004, pp. 689-722.

Hirsch, Barry T. and John V. Winters. 2014. “An Anatomy of Racial and Ethnic Trends in Male Earnings,” Review of Income and Wealth 60(4): 930-47.

Kirkeboen, Lars J., Edwin Leuven, and Magne Mogstad. 2016. “Field of Study, Earnings, and Self-Selection.” Quarterly Journal of Economics 131(3): 1057-1111.

Lewis, Gregory B. and Rahul Pathak. 2014. “The Employment of Veterans in State and Local Government Service,” Public Management & Policy Faculty Publications, 11, Georgia State Univ. https://scholarworks.gsu.edu/pmap_facpubs/11.

Little, Roger D. and J. Eric Fredland. 1979. “Veteran Status, Earnings, and Race: Some Long Term Results,” Armed Forces & Society, 5(2): 244-60.

Mangum, Stephen L. and David E. Ball. 1989. “The Transferability of Military-Provided Occupational Training in the Post-Draft Era,” Industrial and Labor Relations Review 42(2): 230-45.

Makridis, Christos A. 2021. “(Why) Is There a Public/Private Pay Gap?” Journal of Government

and Economics, 1.

23 Martorell, Paco, Trey Miller, and Lindsay Daugherty, and Mark Borgschulte. 2013. Effects of

Military Service on Earnings and Education Revisited: Variation by Service Duration,

Occupation, and Civilian Unemployment. RAND Corporation, Research Report 342.

Mehay, Stephen L. and Barry T. Hirsch. 1996. “The Post-Military Earnings of Female Veterans,” Industrial Relations 35(2): 197-217.

Routan, P. Wesley. 2014. “The Effect of 21st Century Military Service on Civilian Labor and Educational Outcomes,” Journal of Labor Research 35(1): 15-38.

Steinhauer, Jennifer. 2020. “Veterans Are Working, but Not in Jobs That Match Their Advanced Training.” New York Times, Updated 3/16, accessed at https://www.nytimes.com/2020/03/07/us/politics/veterans-jobs-employment.html

Tan, Hui Ren. 2020. “Did Military Service During World War I Affect the Economic Status of American Veterans? Explorations in Economic History 75(1): Article 101301.

Schwartz, Saul. 1986. “The Relative Earnings of Vietnam and Korean-Era Veterans.” Industrial

and Labor Relations Review 39(4): 564-72.

Spence, Michael. 1973. “Job Market Signaling,” Quarterly Journal of Economics 87(3): 355-74.

U.S. Bureau of Labor Statistics. 2019a. Employment Situation of Veterans—2018, USDL-19-

0451, March 21.

U.S. Bureau of Labor Statistic. 2019b. Union Members—2018, USDL-19-0079, January 18.

U.S. Congressional Budget Office. 2017. Labor Force Experiences of Recent Veterans, May, accessed at https://www.cbo.gov/system/files/115th-congress-2017-2018/reports/52418-laborforceveterans_0.pdf

Villemez, W.J. and J.D. Kasarda. 1976. “Veteran Status and Socioeconomic Attainment,” Armed

Forces and Society, Vol. 2: 407-20.

Wenger, Jeffrey B., et al. 2017. Helping Soldiers Leverage Army Knowledge, Skills, and Abilities

in Civilian Jobs, RAND Corporation, Research Report 1719-A.

Wenger, Jeffrey B., et al. 2017. Helping Soldiers Leverage Army Knowledge, Skills, and Abilities

in Civilian Jobs, RAND Corporation, Research Report 1719-A.

Zivin, Kara, Amy S. B. Bohnert, Briana Mezuk, Mark A. Ilgen, Deborah Welsh, Scott Ratliff, Erin M. Miller, Marcia Valenstein, and Amy M. Kilbourne. 2011. “Employment Status of Patients in the VA Health System: Implications for Mental Health Services,” Psychiatric

Services 62(1): 35-38.

Table 1: Male Hourly Earnings Differences Among Veterans and Nonveterans, by Grouplog(Hourly Real Earnings)

(1) (2) (3) (4) (5) (6) (7) (8) (9)College Degree .740∗∗∗ .540∗∗∗ .545∗∗∗ .539∗∗∗ .538∗∗∗ .540∗∗∗ .534∗∗∗ .539∗∗∗

[.002] [.002] [.002] [.002] [.002] [.002] [.002] [.002]Postgraduate Degree .957∗∗∗ .700∗∗∗ .706∗∗∗ .700∗∗∗ .699∗∗∗ .700∗∗∗ .695∗∗∗ .700∗∗∗

[.003] [.003] [.003] [.003] [.003] [.003] [.003] [.003]Is Black -.213∗∗∗ -.154∗∗∗ -.154∗∗∗ -.157∗∗∗ -.154∗∗∗ -.154∗∗∗ -.154∗∗∗ -.155∗∗∗

[.002] [.002] [.002] [.002] [.002] [.002] [.002] [.002]Is Hispanic -.153∗∗∗ -.117∗∗∗ -.116∗∗∗ -.117∗∗∗ -.122∗∗∗ -.116∗∗∗ -.112∗∗∗ -.117∗∗∗

[.002] [.002] [.002] [.002] [.002] [.002] [.002] [.002]Rural Area -.136∗∗∗

[.002]Union Coverage .134∗∗∗

[.002]Public Employee .012∗∗∗

[.003]Is a Veteran .123∗∗∗ .013∗∗∗ .002 .018∗∗∗ -.001 -.005∗∗∗ .007∗∗∗ -.001 -.005∗∗

[.002] [.002] [.002] [.002] [.002] [.002] [.002] [.002] [.002]× Has College Degree -.055∗∗∗

[.004]× Is Black .024∗∗∗

[.005]× Is Hispanic .098∗∗∗

[.007]× Rural Area -.028∗∗∗

[.004]× Union Coverage .016∗∗∗

[.004]× Public Employee .031∗∗∗

[.004]R-squared .00 .35 .42 .42 .42 .42 .42 .43 .42Sample Size 1151288 1151288 1151288 1151288 1151288 1151288 1151288 1151288 1151288Individual Controls No Yes Yes Yes Yes Yes Yes Yes YesOccupation FE No No Yes Yes Yes Yes Yes Yes YesIndustry FE No No Yes Yes Yes Yes Yes Yes YesTime FE No No Yes Yes Yes Yes Yes Yes Yes

Notes.–Sources: Monthly Current Population Survey, 2005-2018. The table reports, restricted to males only, the coefficientsassociated with regressions of logged hourly earnings (deflating nominal weakly earnings using the personal consumptionexpenditure index with 2012 as a base year) on an indicator for veteran status and various interaction effects, controlling for aquadratic in potential experience (age net of years of schooling minus 5), education fixed effects (high school degree, some college,associates, college, and postgraduate, normalized to less than a high school degree), race (white & hispanic, black, normalized towhite & non-hispanic), married, family size, and fixed effects on metropolitan size (over 5 million, 2.49-4.9 million, 1-2.49 million,500,000-999,999, 249,999-499,999, 100,000-249,999, normalized to a non-metro area). Public employees are those working in thefederal, state, or local government. The sample is restricted to males, specifically those who are employed and not in full-timeschool. Standard errors are heteroskedasticity-robust and observations are weighted using earnwt.

Table 2: Robustness in the Current Population Survey Using Female Hourly Earningslog(Hourly Real Earnings)

(1) (2) (3) (4) (5) (6) (7) (8) (9)College Degree .722∗∗∗ .471∗∗∗ .472∗∗∗ .471∗∗∗ .471∗∗∗ .471∗∗∗ .469∗∗∗ .471∗∗∗

[.002] [.002] [.002] [.002] [.002] [.002] [.002] [.002]Postgraduate Degree .946∗∗∗ .660∗∗∗ .661∗∗∗ .660∗∗∗ .660∗∗∗ .660∗∗∗ .653∗∗∗ .659∗∗∗

[.003] [.003] [.003] [.003] [.003] [.003] [.003] [.003]Is Black -.103∗∗∗ -.082∗∗∗ -.082∗∗∗ -.082∗∗∗ -.082∗∗∗ -.082∗∗∗ -.083∗∗∗ -.083∗∗∗

[.002] [.002] [.002] [.002] [.002] [.002] [.002] [.002]Is Hispanic -.103∗∗∗ -.068∗∗∗ -.068∗∗∗ -.068∗∗∗ -.069∗∗∗ -.068∗∗∗ -.067∗∗∗ -.068∗∗∗

[.002] [.002] [.002] [.002] [.002] [.002] [.002] [.002]Rural Area -.166∗∗∗

[.002]Union Coverage .110∗∗∗

[.002]Public Employee .025∗∗∗

[.002]Is a Veteran .143∗∗∗ .044∗∗∗ .004 .017∗∗∗ .004 -.000 .007 .003 -.008

[.006] [.005] [.005] [.005] [.005] [.005] [.005] [.005] [.006]× Has College Degree -.033∗∗∗

[.010]× Is Black .001

[.011]× Is Hispanic .063∗∗∗

[.019]× Rural Area -.012

[.010]× Union Coverage .008

[.013]× Public Employee .039∗∗∗

[.010]R-squared .00 .32 .41 .41 .41 .41 .41 .41 .41Sample Size 1106956 1106956 1106956 1106956 1106956 1106956 1106956 1106956 1106956Individual Controls No Yes Yes Yes Yes Yes Yes Yes YesOccupation FE No No Yes Yes Yes Yes Yes Yes YesIndustry FE No No Yes Yes Yes Yes Yes Yes YesTime FE No No Yes Yes Yes Yes Yes Yes Yes

Notes.–Sources: Monthly Current Population Survey, 2005-2018. The table reports, restricted to females only, the coefficientsassociated with regressions of logged hourly earnings (deflating nominal weakly earnings using the personal consumptionexpenditure index with 2012 as a base year) on an indicator for veteran status and various interaction effects, controlling for aquadratic in potential experience (age net of years of schooling minus 5), education fixed effects (high school degree, some college,associates, college, and postgraduate, normalized to less than a high school degree), race (white & hispanic, black, normalized towhite & non-hispanic), married, family size, and fixed effects on metropolitan size (over 5 million, 2.49-4.9 million, 1-2.49 million,500,000-999,999, 249,999-499,999, 100,000-249,999, normalized to a non-metro area). Public employees are those working in thefederal, state, or local government. The sample is restricted to males, specifically those who are employed and not in full-timeschool. Standard errors are heteroskedasticity-robust and observations are weighted using earnwt.

Tab

le3:

Rob

ust

nes

sU

sing

the

Am

eric

anC

omm

unit

ySurv

eyfo

rA

nnual

Ear

nin

gsD

iffer

ence

slo

g(A

nnual

Rea

lE

arnin

gs)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

(14)

Isa

Vet

eran

.073

∗∗∗

-.01

7∗∗∗

-.04

6∗∗∗

-.02

8∗∗∗

-.00

5∗∗∗

-.03

1∗∗∗

-.03

2∗∗∗

.217

∗∗∗

.090

∗∗∗

-.00

3-.

016∗

∗∗

.006

∗∗

-.01

8∗∗∗

-.01

8∗∗∗

[.00

1][.00

1][.00

1][.00

1][.00

1][.00

1][.00

1][.00

3][.00

3][.00

3][.00

2][.00

3][.00

2][.00

2]×

Has

Col

lege

Deg

ree

-.07

9∗∗∗

-.05

9∗∗∗

[.00

2][.00

5]×

IsB

lack

.033

∗∗∗

.011

∗

[.00

3][.00

6]×

IsH

ispan

ic.0

82∗∗∗

.044

∗∗∗

[.00

4][.01

1]lo

g(H

ours

Wor

ked

per

Wee

k)

.863

∗∗∗

.862

∗∗∗

.863

∗∗∗

.863

∗∗∗

.929

∗∗∗

.929

∗∗∗

.929

∗∗∗

.929

∗∗∗

[.00

2][.00

2][.00

2][.00

2][.00

1][.00

1][.00

1][.00

1]14

-26

Wee

ks

Las

tY

ear

.851

∗∗∗

.851

∗∗∗

.851

∗∗∗

.851

∗∗∗

.875

∗∗∗

.875

∗∗∗

.875

∗∗∗

.875

∗∗∗

[.00

4][.00

4][.00

4][.00

4][.00

4][.00

4][.00

4][.00

4]27

-39

Wee

ks

Las

tY

ear

1.29

9∗∗∗

1.29

9∗∗∗

1.29

9∗∗∗

1.29

9∗∗∗

1.35

6∗∗∗

1.35

6∗∗∗

1.35

6∗∗∗

1.35

6∗∗∗

[.00

4][.00

4][.00

4][.00

4][.00

4][.00

4][.00

4][.00

4]40

-47

Wee

ks

Las

tY

ear

1.60

5∗∗∗

1.60

5∗∗∗

1.60

5∗∗∗

1.60

5∗∗∗

1.64

3∗∗∗

1.64

3∗∗∗

1.64

3∗∗∗

1.64

3∗∗∗

[.00

4][.00

4][.00

4][.00

4][.00

3][.00

3][.00

3][.00

3]48

-49

Wee

ks

Las

tY

ear

1.77

1∗∗∗

1.77

0∗∗∗

1.77

0∗∗∗

1.77

1∗∗∗

1.76

1∗∗∗

1.76

1∗∗∗

1.76

1∗∗∗

1.76

1∗∗∗

[.00

4][.00

4][.00

4][.00

4][.00

4][.00

4][.00

4][.00

4]50

-52

Wee

ks

Las

tY

ear

1.86

7∗∗∗

1.86

7∗∗∗

1.86

7∗∗∗

1.86

7∗∗∗

1.87

0∗∗∗

1.87

0∗∗∗

1.87

0∗∗∗

1.87

0∗∗∗

[.00

4][.00

4][.00

4][.00

4][.00

3][.00

3][.00

3][.00

3]R

-squar

ed.0

0.2

9.3

9.6

1.6

1.6

1.6

1.0

0.2

0.3

3.6

6.6

6.6

6.6

6Sam

ple

Siz

e84

1966

584

1966

577

6858

377

6858

377

6858

377

6858

377

6858

375

9194

175

9194

170

0491

370

0491

370

0491

370

0491

370

0491

3In

div

idual

Con

trol

sN

oY

esY

esY

esY

esY

esY

esN

oY

esY

esY

esY

esY

esY

esO

ccupat

ion

FE

No

No

Yes

Yes

Yes

Yes

Yes

No

No

Yes

Yes

Yes

Yes

Yes

Indust

ryF

EN

oN

oY

esY

esY

esY

esY

esN

oN

oY

esY

esY

esY

esY

esT

ime

FE

No

No

Yes

Yes

Yes

Yes

Yes

No

No

Yes

Yes

Yes

Yes

Yes

Sam

ple

Mal

eM

ale

Mal

eM

ale

Mal

eM

ale

Mal

eFem

ale

Fem

ale

Fem

ale

Fem

ale

Fem

ale

Fem

ale

Fem

ale

Notes.–

Sou

rces

:A

mer

ican

Com

mu

nit

yS

urv

ey,

2005-2

018.

Th

eta

ble

rep

ort

s,se

para

tely

by

gen

der

,th

eco

effici

ents

ass

oci

ate

dw

ith

regre

ssio

ns

of

logged

an

nu

al

earn

ings

(defl

ati

ng

nom

inal

wea

kly

earn

ings

usi

ng

the

per

son

al

con

sum

pti

on

exp

end

itu

rein

dex

wit

h2012

as

ab

ase

yea

r)on

an

ind

icato

rfo

rvet

eran

statu

s,co

ntr

oll

ing

for

aq

uad

rati

cin

pote

nti

al

exp

erie

nce

(age

net

of

yea

rsof

sch

ooli

ng

min

us

5),

edu

cati

on

fix

edeff

ects

(hig

hsc

ho

ol

deg

ree,

som

eco

lleg

e,ass

oci

ate

s,co

lleg

e,an

dp

ost

gra

du

ate

,n

orm

ali

zed

tole

ssth

an

ah

igh

sch

ool

deg

ree)

,ra

ce(w

hit

e&

his

pan

ic,

bla

ck,

norm

ali

zed

tow

hit

e&

non

-his

pan

ic),

an

da

marr

ied

ind

icato

r.T

he

sam

ple

isre

stri

cted

toth

ose

wh

oare

emp

loyed

an

dn

ot

infu

ll-t

ime

sch

ool.

Sta

nd

ard

erro

rsare

het

erosk

edast

icit

y-r

ob

ust

an

dob

serv

ati

on

sare

wei

gh

ted

usi

ng

perwt.

.04

.06

.08

.1.1

2Ve

tera

ns S

hare

in O

ccup

atio

n (2

010-

2017

Ave

rage

)

-.1 0 .1 .2 .3Annual Real Hourly Wage (2002-2018 Growth)

Employment-weighted Correlation = -0.12

Panel A: Veterans Share and Hourly Wage Growth

.04

.06

.08

.1.1

2Ve

tera

ns S

hare

in O

ccup

atio

n (2

010-

2017

Ave

rage

)

-.5 0 .5 1 1.5Annual Employment (2002-2018 Growth)

Employment-weighted Correlation = -0.12

Panel B: Veterans Share and Employment Growth

Figure 1: Earnings and Employment Growth Across OccupationsNotes.–Sources: American Community Survey, 2010-2017, Occupational Employment Statistics, 2002-2018. The figure plots the shareof veterans at a four-digit SOC level from 2010-2017 with the growth in employment and real hourly wages (deflated with 2012 pricesfrom the personal consumption expenditure index) between the 2002-03 and 2017-18 averages. Observations are weighted byemployment in the four-digit occupational cell.

.06

.07

.08

.09

.1Ve

tera

ns S

hare

in O

ccup

atio

n (2

010-

2017

Ave

rage

)

-.2 0 .2 .4Annual Real GDP/capita Growth (2002-2017 Growth)

Employment-weighted Correlation = -0.37

Panel A: Veterans Share and Real GDP/capita Growth

.05

.06

.07

.08

.09

.1Ve

tera

ns in

Occ

upat

ion

(201

0-20

17 A

vera

ge)

10 10.5 11 11.5log(Real GDP/capita) (2002-2017 Growth)

Employment-weighted Correlation = -0.46

Panel B: Veterans Share and log(Real GDP/capita)

Figure 2: Real GDP and Veterans Incidence Across Metropolitan AreasNotes.–Sources: American Community Survey, 2010-2017, Bureau of Economic Analysis, 2002-2018. The figure plots the share ofveterans at a four-digit SOC level from 2010-2017 with logged real (in 2012 prices) GDP per capita 2017-18 average and its growthbetween the 2002-03 and 2017-18 averages. Observations are weighted by the number of respondents from the ACS in a givenmetropolitan area.

-.1-.0

50

.05

.1.1

5Li

cens

ing

for V

eter

ans

Manag

emen

t/bus

iness

Math/C

S/engin

eerin

g

Commun

ity/ed

ucati

on/ar

ts

Health

/prac

tition

er

Service

Sales/a

dmin

Constr

uctio

n/prod

uctio

n/tran

sport

Occupation

Figure 3: Heterogeneity in Earnings Among Veterans With versus Without Certifications/LicensesNotes.–Sources: CPS Monthly, 2015-2019. The figure reports the coefficients associated with regressions of logged earnings on anindicator for whether the individual is a veteran, an indicator for whether the individual has a professional license (excludingbusiness), and the interaction, conditional on controls, separately for major occupational groups.

0 .05 .1 .15 .2 .25percent

Other

Industrial

Medicine

Communications

Sciences

Humanities

Education

Engineering/CS

Social sciences

Business

Non-Veteran Veteran

Figure 4: Distribution of College Degrees for Veterans and NonveteransNotes.–Sources: ACS, 2009-2017. The figure plots the share of individuals with different types of college degrees.

Table 6: Labor Market Returns to Different College Degrees, Veterans and Nonveteranslog(Annual Earnings)

(1) (2) (3)Communications .098∗∗∗ .125∗∗∗ .019∗∗∗

[.002] [.002] [.002]Engineering/CS .423∗∗∗ .265∗∗∗ .105∗∗∗

[.002] [.002] [.002]Education -.267∗∗∗ -.227∗∗∗ -.067∗∗∗

[.002] [.002] [.002]Humanities -.179∗∗∗ -.145∗∗∗ -.065∗∗∗

[.002] [.002] [.002]Sciences .289∗∗∗ .134∗∗∗ .025∗∗∗

[.003] [.002] [.002]Industrial/Production .042∗∗∗ -.024∗∗∗ -.016∗∗∗

[.004] [.003] [.003]Other -.079∗∗∗ -.069∗∗∗ -.023∗∗∗

[.004] [.004] [.004]Medicine .093∗∗∗ .184∗∗∗ .047∗∗∗

[.002] [.002] [.003]Business/Law .153∗∗∗ .156∗∗∗ .059∗∗∗

[.002] [.002] [.002]Veteran .135∗∗∗ -.018∗∗∗ -.035∗∗∗

[.005] [.005] [.005]× Communications .013 -.034∗∗∗ -.017∗∗

[.010] [.009] [.008]× Engineering/CS -.135∗∗∗ -.012 .015∗∗

[.008] [.008] [.007]× Education -.064∗∗∗ -.058∗∗∗ -.068∗∗∗

[.011] [.010] [.010]× Humanities -.000 -.009 .010

[.010] [.010] [.009]× Sciences .036∗∗∗ .106∗∗∗ .047∗∗∗

[.010] [.010] [.009]× Industrial/Production -.007 .055∗∗∗ .043∗∗∗

[.013] [.013] [.011]× Other -.045∗ -.003 -.015

[.023] [.023] [.021]× Medicine .046∗∗∗ .028∗∗∗ .021∗∗

[.011] [.010] [.009]× Business/Law -.129∗∗∗ -.083∗∗∗ -.040∗∗∗

[.007] [.007] [.006]R-squared .04 .19 .35Sample Size 4192985 4192985 4192985Individual Controls No Yes YesOccupation FE No No YesYear FE No No Yes

Notes.–Sources: American Community Survey, 2009-2017. The table reports the coefficients associated with regressions of loggedannual earnings (deflated with the 2012 personal consumption expenditure deflator) on an indicator for being a veteran, anindicator for whether the individual has undergraduate college degree d (where d denotes the columns), and their interaction,conditional on controls. These controls include: a quadratic of age, marital status, race (white, black, asian), years of schooling,and four-digit occupational and year fixed effects. Standard errors are heteroskedasticity-robust and observations are weighted bythe sample weights.